2 * Functions related to segment and merge handling
4 #include <linux/kernel.h>
5 #include <linux/module.h>
7 #include <linux/blkdev.h>
8 #include <linux/scatterlist.h>
12 static unsigned int __blk_recalc_rq_segments(struct request_queue *q,
15 struct bio_vec *bv, *bvprv = NULL;
16 int cluster, i, high, highprv = 1;
17 unsigned int seg_size, nr_phys_segs;
18 struct bio *fbio, *bbio;
24 cluster = blk_queue_cluster(q);
28 bio_for_each_segment(bv, bio, i) {
30 * the trick here is making sure that a high page is
31 * never considered part of another segment, since that
32 * might change with the bounce page.
34 high = page_to_pfn(bv->bv_page) > queue_bounce_pfn(q);
38 if (seg_size + bv->bv_len
39 > queue_max_segment_size(q))
41 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
43 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
46 seg_size += bv->bv_len;
51 if (nr_phys_segs == 1 && seg_size >
52 fbio->bi_seg_front_size)
53 fbio->bi_seg_front_size = seg_size;
57 seg_size = bv->bv_len;
63 if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size)
64 fbio->bi_seg_front_size = seg_size;
65 if (seg_size > bbio->bi_seg_back_size)
66 bbio->bi_seg_back_size = seg_size;
71 void blk_recalc_rq_segments(struct request *rq)
73 rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio);
76 void blk_recount_segments(struct request_queue *q, struct bio *bio)
78 struct bio *nxt = bio->bi_next;
81 bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio);
83 bio->bi_flags |= (1 << BIO_SEG_VALID);
85 EXPORT_SYMBOL(blk_recount_segments);
87 static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
90 if (!blk_queue_cluster(q))
93 if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
94 queue_max_segment_size(q))
97 if (!bio_has_data(bio))
100 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
104 * bio and nxt are contiguous in memory; check if the queue allows
105 * these two to be merged into one
107 if (BIO_SEG_BOUNDARY(q, bio, nxt))
114 * map a request to scatterlist, return number of sg entries setup. Caller
115 * must make sure sg can hold rq->nr_phys_segments entries
117 int blk_rq_map_sg(struct request_queue *q, struct request *rq,
118 struct scatterlist *sglist)
120 struct bio_vec *bvec, *bvprv;
121 struct req_iterator iter;
122 struct scatterlist *sg;
126 cluster = blk_queue_cluster(q);
133 rq_for_each_segment(bvec, rq, iter) {
134 int nbytes = bvec->bv_len;
136 if (bvprv && cluster) {
137 if (sg->length + nbytes > queue_max_segment_size(q))
140 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
142 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
145 sg->length += nbytes;
152 * If the driver previously mapped a shorter
153 * list, we could see a termination bit
154 * prematurely unless it fully inits the sg
155 * table on each mapping. We KNOW that there
156 * must be more entries here or the driver
157 * would be buggy, so force clear the
158 * termination bit to avoid doing a full
159 * sg_init_table() in drivers for each command.
161 sg->page_link &= ~0x02;
165 sg_set_page(sg, bvec->bv_page, nbytes, bvec->bv_offset);
169 } /* segments in rq */
172 if (unlikely(rq->cmd_flags & REQ_COPY_USER) &&
173 (blk_rq_bytes(rq) & q->dma_pad_mask)) {
174 unsigned int pad_len =
175 (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
177 sg->length += pad_len;
178 rq->extra_len += pad_len;
181 if (q->dma_drain_size && q->dma_drain_needed(rq)) {
182 if (rq->cmd_flags & REQ_WRITE)
183 memset(q->dma_drain_buffer, 0, q->dma_drain_size);
185 sg->page_link &= ~0x02;
187 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
189 ((unsigned long)q->dma_drain_buffer) &
192 rq->extra_len += q->dma_drain_size;
200 EXPORT_SYMBOL(blk_rq_map_sg);
202 static inline int ll_new_hw_segment(struct request_queue *q,
206 int nr_phys_segs = bio_phys_segments(q, bio);
208 if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q))
211 if (bio_integrity(bio) && blk_integrity_merge_bio(q, req, bio))
215 * This will form the start of a new hw segment. Bump both
218 req->nr_phys_segments += nr_phys_segs;
222 req->cmd_flags |= REQ_NOMERGE;
223 if (req == q->last_merge)
224 q->last_merge = NULL;
228 int ll_back_merge_fn(struct request_queue *q, struct request *req,
231 unsigned short max_sectors;
233 if (unlikely(req->cmd_type == REQ_TYPE_BLOCK_PC))
234 max_sectors = queue_max_hw_sectors(q);
236 max_sectors = queue_max_sectors(q);
238 if (blk_rq_sectors(req) + bio_sectors(bio) > max_sectors) {
239 req->cmd_flags |= REQ_NOMERGE;
240 if (req == q->last_merge)
241 q->last_merge = NULL;
244 if (!bio_flagged(req->biotail, BIO_SEG_VALID))
245 blk_recount_segments(q, req->biotail);
246 if (!bio_flagged(bio, BIO_SEG_VALID))
247 blk_recount_segments(q, bio);
249 return ll_new_hw_segment(q, req, bio);
252 int ll_front_merge_fn(struct request_queue *q, struct request *req,
255 unsigned short max_sectors;
257 if (unlikely(req->cmd_type == REQ_TYPE_BLOCK_PC))
258 max_sectors = queue_max_hw_sectors(q);
260 max_sectors = queue_max_sectors(q);
263 if (blk_rq_sectors(req) + bio_sectors(bio) > max_sectors) {
264 req->cmd_flags |= REQ_NOMERGE;
265 if (req == q->last_merge)
266 q->last_merge = NULL;
269 if (!bio_flagged(bio, BIO_SEG_VALID))
270 blk_recount_segments(q, bio);
271 if (!bio_flagged(req->bio, BIO_SEG_VALID))
272 blk_recount_segments(q, req->bio);
274 return ll_new_hw_segment(q, req, bio);
277 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
278 struct request *next)
280 int total_phys_segments;
281 unsigned int seg_size =
282 req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
285 * First check if the either of the requests are re-queued
286 * requests. Can't merge them if they are.
288 if (req->special || next->special)
292 * Will it become too large?
294 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > queue_max_sectors(q))
297 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
298 if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
299 if (req->nr_phys_segments == 1)
300 req->bio->bi_seg_front_size = seg_size;
301 if (next->nr_phys_segments == 1)
302 next->biotail->bi_seg_back_size = seg_size;
303 total_phys_segments--;
306 if (total_phys_segments > queue_max_segments(q))
309 if (blk_integrity_rq(req) && blk_integrity_merge_rq(q, req, next))
313 req->nr_phys_segments = total_phys_segments;
318 * blk_rq_set_mixed_merge - mark a request as mixed merge
319 * @rq: request to mark as mixed merge
322 * @rq is about to be mixed merged. Make sure the attributes
323 * which can be mixed are set in each bio and mark @rq as mixed
326 void blk_rq_set_mixed_merge(struct request *rq)
328 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
331 if (rq->cmd_flags & REQ_MIXED_MERGE)
335 * @rq will no longer represent mixable attributes for all the
336 * contained bios. It will just track those of the first one.
337 * Distributes the attributs to each bio.
339 for (bio = rq->bio; bio; bio = bio->bi_next) {
340 WARN_ON_ONCE((bio->bi_rw & REQ_FAILFAST_MASK) &&
341 (bio->bi_rw & REQ_FAILFAST_MASK) != ff);
344 rq->cmd_flags |= REQ_MIXED_MERGE;
347 static void blk_account_io_merge(struct request *req)
349 if (blk_do_io_stat(req)) {
350 struct hd_struct *part;
353 cpu = part_stat_lock();
354 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
356 part_round_stats(cpu, part);
357 part_dec_in_flight(part, rq_data_dir(req));
364 * Has to be called with the request spinlock acquired
366 static int attempt_merge(struct request_queue *q, struct request *req,
367 struct request *next)
369 if (!rq_mergeable(req) || !rq_mergeable(next))
373 * Don't merge file system requests and discard requests
375 if ((req->cmd_flags & REQ_DISCARD) != (next->cmd_flags & REQ_DISCARD))
379 * Don't merge discard requests and secure discard requests
381 if ((req->cmd_flags & REQ_SECURE) != (next->cmd_flags & REQ_SECURE))
387 if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next))
390 if (rq_data_dir(req) != rq_data_dir(next)
391 || req->rq_disk != next->rq_disk
396 * If we are allowed to merge, then append bio list
397 * from next to rq and release next. merge_requests_fn
398 * will have updated segment counts, update sector
401 if (!ll_merge_requests_fn(q, req, next))
405 * If failfast settings disagree or any of the two is already
406 * a mixed merge, mark both as mixed before proceeding. This
407 * makes sure that all involved bios have mixable attributes
410 if ((req->cmd_flags | next->cmd_flags) & REQ_MIXED_MERGE ||
411 (req->cmd_flags & REQ_FAILFAST_MASK) !=
412 (next->cmd_flags & REQ_FAILFAST_MASK)) {
413 blk_rq_set_mixed_merge(req);
414 blk_rq_set_mixed_merge(next);
418 * At this point we have either done a back merge
419 * or front merge. We need the smaller start_time of
420 * the merged requests to be the current request
421 * for accounting purposes.
423 if (time_after(req->start_time, next->start_time))
424 req->start_time = next->start_time;
426 req->biotail->bi_next = next->bio;
427 req->biotail = next->biotail;
429 req->__data_len += blk_rq_bytes(next);
431 elv_merge_requests(q, req, next);
434 * 'next' is going away, so update stats accordingly
436 blk_account_io_merge(next);
438 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
439 if (blk_rq_cpu_valid(next))
440 req->cpu = next->cpu;
442 /* owner-ship of bio passed from next to req */
444 __blk_put_request(q, next);
448 int attempt_back_merge(struct request_queue *q, struct request *rq)
450 struct request *next = elv_latter_request(q, rq);
453 return attempt_merge(q, rq, next);
458 int attempt_front_merge(struct request_queue *q, struct request *rq)
460 struct request *prev = elv_former_request(q, rq);
463 return attempt_merge(q, prev, rq);